support dif between two coverage run

While I did suggest to someone (who I assume is i928) that he post here, he contacted me on his own, asking about Tripoli (which does differential code coverage, as opposed to incremental code coverage). I had been meaning to ask for a differential code coverage option as well.

I think incremental code coverage is easier to understand and explain. However, I have been finding that I frequently want differential code coverage instead. One anecdote: I was working on a blog posting about incremental code coverage (due to come out on Monday, BTW) and loaded up the code to jHotdraw, which I had used in a user study a few years back. Using differential code coverage, users were able to isolate {the code specific to adding arrowheads} down to only a very small handful of classes. However, when I tried it with incremental code coverage, there were lots -- about ten times as many. All the code involved in refreshing the canvas showed up. I ended up needing to go find another example.

I wrote some stupid shell scripts to give me differential code coverage -- I dump the CSV and HTML, diff the CSV to find which HTML files to compare, then diff those -- but it would be much nicer to have in the code.

To make Tripoli, I modified the EclEmma 1.0 Merge code so that instead of doing an "addition", it did a "subtraction". While I haven't looked at the 2.0 code, I imagine something similar would not be difficult.

One possible way to implement this is a improved version of the "Merge" Dialog:

1) For each session in the list you can decide whether the execution data should be added or subtracted.
2) A new checkbox determines whether the merged sessions will be removed (currently merged sessions are always removed)