Merged. Thanks again.

I think the code can be merged.

Please merge it once you both like it. (If you don't think it's necessary, I
won't look at the code or detailed results.)

Forgot to add a link to the latest report:

http://ai.cs.unibas.ch/_tmp_files/pommeren/issue528-base-issue528-v3-compare.html

Thanks Jendrik! I included your suggestions and ran the experiment again. One
node on the grid crashed during the experiment (ase08) so there are 14 runs that
did not finish. The remaining runs confirm the results from the last experiment:
the patched version is about 10% faster than the baseline and solves 5
additional tasks.

Can I merge this?

Done with the review.

Yes, I will do a review.

I forgot to create a pull request on bitbucket, so we can comment on the code.
Here it is:

https://bitbucket.org/flogo/downward-issues-issue528/pull-request/1/issue528/diff

I have a patch ready for review at
https://bitbucket.org/flogo/downward-issues-issue528/branch/issue528#diff

The branch has 2 main commits: the first (v1) changes LM-cut to use the task
interface and the second (v2) does some drive-by cleanup and changes for loops
in LM-cut to range-based loops.

I ran experiments for both versions and the results look good (+5 coverage):

Base -> v1
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue528-base-issue528-v1-compare.html
v1 -> v2
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue528-v1-issue528-v2-compare.html
Base -> v2
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue528-base-issue528-v2-compare.html
Scatter plots for memory and time:
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue528_base_v2_total_time.png
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue528_base_v2_memory.png

Jendrik, could you do a review?

LM-cut should use the task it get from the options object (as a TaskProxy)
instead of the global task.

Once this is done, we can continue with issue527.