Preamble: 

The paper was accepted as a short paper for CSMR 2009 (but submitted
as a full paper of course). We decided to turn down the offer for a
short paper because (i) the paper wasn't meant as a short paper; (ii)
we couldn't decide on how to derive a useful short paper; (iii) we
think that we ended up in the short-paper category for unfortunate
reasons -- a) review 1 is quite superficial (if not content-free) in
our humble opinion; b) review 3 recommended rejection solely on the
grounds that the "introduction to grammar convergence" would need to
be published prior to the publication of this paper -- which it
meanwhile is.

Conclusion: the best course of action seemed to be to waive the offer
for a short paper and to improve the paper anyway.

Ralf Laemmel
8.2.2009

------------------------------------------------------------

Changes compared to the CSMR 2009 submission.

Overall this new version is a full rewrite not so much based on
detailed reviewer feedback, but rather because 4 additional months
went into the project, and the overall sound of the CSMR notification
showed us that we need to explain and illustrate some parts just way
better.

The title has changed, too: the old title was a pun (an insider joke)
that was explained by the footnote that was attached to it. Still, two
reviewers struggled, and so we gave up on this pun.

The ratio extraction/transformation was changed in favor of
transformation: we agree with CSMR reviewer 2 that the extraction part
is certainly not the part with the strongest impact and strongest
level of originality. Accordingly, we have increased the level of
detail on the transformation approach and downscaled the discussion of
the extraction approach while we preserved what we think is an
original design of a non-classic parser that may also be helpful for
other extraction efforts. We also used 2 additional pages for details
on transformations.

Added details of transformation language: we categorize the kind of
transformations (refactoring, optimization, extension, correction,
recovery), and illustrate all categories with appropriate samples of
operator applications.

Introduction and conclusion: both have been written up from scratch to
better explain the impact/the importance of this work and the overall
approach. In the conclusion, we also clarify what's missing and hence
left for future work.

Related work: since at some level of abstraction, this Java effort is
comparable to our earlier work on grammar recovery (mainly Cobol), but
also other people's work on grammar recovery, we have done a better
job at explaining the relationship. We have also explained how grammar
comparison relates to similar operations in the ER/relational and
model-driven communities -- the main insight is here that we have not
yet used any sophisticated matching, but this may be the central tool
to bring more automation (inference) to grammar convergence.