We now use sqljdbc4. (note the suffix 4)

  between successive runs

This should make the code a somewhat clearer

Given dependencies on test-jar logback could not be build as 'mvn package' from 
a pristine local repository. One had to 'mvn install' first. 

- doing integration tests with various databases for logback-access

- minor refactoring in DBAppender and related classes
- As suggested by John Gibson in LBCORE-15, we now use the prepareStatement method of 
  Connection which takes the sql statement followed by the autogenerateed key columnNames 
  (of type String[])

  This change is compatible with JDBC drivers of SQLServer and MySQL, as well as Oracle.

  JDBC drivers not supporting getGeneratedKeys are not affected by this change.

- added integration tests which are active only on my machine, namely, Orion. This avoids other
  developers from having to install Oracle, MySQL, Postgres and SQL Server just for the 
  logback tests to pass.
  
- similar tests on logback-access to follow.  

- detachAppender in AppenderAttachable now returns a boolean instead of an Appender.
- iteratorForAppenders() in AppenderAttachable now returns Iterator<Appender<E>> instead of just Iterator
- modified Logger to invoke callAppender without synchronization on this
- AppenderAttachableImpl modified to take advantage of the fact that reads occur much more 
  often than writes
- associated test cases

- renamed ClassPackagingDataCalculator as PackagingDataCalculator
- added calculatePackagingData method to ThrowableProxy so that converters can
  force calculation of packaging data
- added relevant test cases

- Fixes both LBCORE-27 and LBCLASSIC-67. 

- added test cases (using GreenMail) that verify that the output generated
  by SMTPAppender and HTMLLayout conform to xhtml1-strict.dtd

- HTMLLayout* now use StringBuilder instead of StringBuffer. The former is significantly faster.

Other missing files

Not not all PatternLayout instances need to have a
ThrowableInformationConverter added at the end of the converter chain
by the postCompileProcessing() method.

With this commit, post compile processing code has been moved to an
interface. By default, PatternLayouts that ship with logback have a
default post compile processor. However, it is now very easy to set
processor to null.

We now test SMTPAppender using GreenMail. Problem described in
LBCLASSIC-67 is reproduced in tests but not yet fixed.

Given that GPL+classpath exception is rejected in the same way as LGPL
by those requesting for a licensing change, there is no need for
change if the end result is the same.

  Tomcat), will now systematically print its internal status upon
  initialization, unless told to be quiet. This greatly helps
  troubleshooting the configration of logback-access under Tomcat.

- dropped retrotranslator support

- additional assertion in JettyBasicTest

  getResponseContent after serialization. The idea is, as expalined in comment 3, to make a 
  copy of the contents of various httpRequest and httpResponse fields (without serializing 
  httpRequest and httpResponse). Also note many converters in the c.q.l.acces.Pattern package 
  *cannot* deal with null values returned by various get methods in AccessEvent. Thus, it is not
  enough to avoid NPE as accomplished by attachment 21 (dated 2008-01-28 21:39:11) but also to
  return valid values.

Comment 3: http://bugzilla.qos.ch/show_bug.cgi?id=109#c3

The defensive code compensating for the bug in 0.9.8 until 0.9.9 is released was not merged because it is a workaround hiding the true logic of the code.
 
  void buildResponseHeaderMap() {
    if (responseHeaderMap == null) {
      if(serverAdapter != null) {     <-- ---- not merged
        responseHeaderMap = serverAdapter.buildResponseHeaderMap();
      } else {
        responseHeaderMap = new HashMap<String, String>();
      }
     }
   }

Now, only Context's have a property map. A number of classes has to change to reflect this change.

- first-step to using JUnit 4.