[MKSearch-dev] Week 6 round up

[Phil Shaw]

I'm tending to focus on my frustration with problems in recent days 
rather than some important break-throughs earlier this week, so this 
should help put things in perspective... if I read it from back to 
front.

I have done a half-day catch-up from last week and a half-day extra 
this morning.

Phil

Monday
------

Spent most of the day tracing the problem Bruno reported with JTidy 
and ultimately identified it as libgcj bug 9866. Checked in a set of 
test scripts and an amended version of the ParserImpl class to work 
around the problem. Also cleaned-up a number of early test classes 
that are no longer relevant.

Tuesday
-------

Bruno reported a further problem with TJidy that was caused by the 
same libgcj bug 9866, so this time it was much easier to identify and 
correct. Checked in a minimal test case document for the problem, a 
test script and amendment to work around.

Stripped the test configuration for JSpider to a bare minimum, 
removing dependencies on Apache packages and completed a test run 
using the Sun runtime. Started a static site for more limited 
testing. Chris and Bruno set up the test host and provided remote 
access for me.

Corrected some compilation script errors caused by the clean-up on 
Monday. All scripts and test cases running successfully.

Wednesday
---------

Started investigating a problem running JSpider with gij where the 
dynamic loading of plugin classes was throwing an 
IncompatibleClassChangeError. The JSpider code is pretty dense and 
poorly documented in this area so required some extensive debugging 
code to track down. Ultimately identified the source of the problem, 
but could not create a simple test case to reproduce it.

Separated test classes for GCJ issues from the standard MKSearch test 
directory. Created a simple test case to verify thread support in gij.

Checked in some unit test and coverage classes for the SourceLister 
components. Checked in the original Perl work on MKSearch.

Thursday
--------

Implemented a temporary workaround to cut-short the plugin loader  
issue with JSpider. That was successful, but revealed another problem 
that I traced to loading the rule sets, it may be a similar cause. In 
this case, instantiating the Rule class throws a 
NegativeArraySizeException, which appears to be internal to the 
libgcj Constructor class.