[Demetercop-info] ant errors

[Jeffrey P. <jd...@cs...>]

I've looked at all the errors in the Jakarta Ant Project and found 17 
that map directly to a point in source code.  Here's what I (and the 
eclipse system) found:

ID 
SEMANTIC 
CODE 
LoD 
Bad Soln
624 
Y 
		?
1125 
		Y	N
1208 
		Y	N
1254 
Y 
		N
1543 
		Y	N
1681 
Y 
		N
1848 
		Y	N
3024 
Y 
		N
3271 
		Y	?
3556 
Y 
		Y
4008 
		Y	?
4366 
Y 
		N
4870 
		Y	?
5056 
		Y	N
5230 
		Y	N
7538 
Y 
		?
8575 
		Y	?
9024 
		Y	?
9574 
		Y	?
9762 
		Y	N
10456 
		Y	Y	Y
10576 
		Y	?

Key:
   ID        ::= Error identier
   SEMANTIC  ::=	Y if a semantic error, N otherwise
   CODE      ::= Y if a coding error, N otherwise
   LoD       ::= Y if caused by LoD violation, ? is don't know yet
   Bad Soln  ::= Y if the solution contained LoD violation

I've checked all the error reports and fixes as well as this file in CVS 
under demetercop/doc/ant-errors.

These results suggest no correlation between LoD violations and errors, 
but I'm looking elsewhere for evidence of this.  I seem to think that 
errors in programs such as parsers and other text tools would contain a 
high percentage of errors that map to source code caused by violating 
LoD... I'll keep looking.

Otherwise, the eclipse system seems to find all errors -- there are 
still a few bugs.  It fixes those errors caused by explicit message 
passing (i.e. method calling), but doesn't fix implicit messages (i.e. 
field references).  Fully implementing this should take a little time, 
but should be a bounded amount of time.

I'd like to get some bug data supporting the hypothesis that violations 
of LoD directly lead to errors in code.. I think this could be a paper 
at AOSD, so I'll concentrate on this for the time being.

If (and, hopefully when) I find data supporting this, I can use some of 
the ideas from the Maryland paper Karl sent regarding statistical 
significance, and I'd probably expand that to include some notion of 
distance between the error and the actual violation.  I.e., the weight 
of the violation would involve a distance (calculated from its position 
in the source code) to the error.

Jeff
-- 
Jeffrey Palm --> http://jeffpalm.com