Thank you for the help everyone, this one is pushed.

Ok. Florian, the code is ready, it also includes the changed includes :-)

I only looked at ms_construction_time, and I think there's no clear negative
trend anymore (e.g., look at the domain summaries before/after, which include
delta sums over all instances). So I'm fine with merging this if Florian had no
objections to the code.

The new experiments look a bit better, search time is "less worse" than before.
I also included "actual_search_time" as an attribute -- there is no visible
difference in search time, however, everything is related to merge-and-shrink
construction time.

http://ai.cs.unibas.ch/_tmp_files/sieverss/issue548-base-v2-issue548-base-issue548-v2-compare.html

That's a very good question indeed. As I changed several places with using
tasks/task proxies again after the first experiment, I'd better run another set
of experiments so the data belongs to the most recent version.

Interestingly, the difference in heuristic construction time is considerable. I
find that surprising: one we've created the atomic abstractions, shouldn't the
original task representation be irrelevant?

I updated the pull request.

The following table now also includes "ms_construction_time" (time included in
search time/total time, which is spent on the merge-and-shrink construction):
http://ai.cs.unibas.ch/_tmp_files/sieverss/issue548-mas-refetch-issue548-base-issue548-v1-compare.html

Actually, I find runtime a bit disappointing. Is it possible to find out how
much is actual search time and how much is the time for constructing the heuristic?

Great! Looks like Florian is reviewing this, so please merge once you're both
happy with it.

The experiments look ok:
http://ai.cs.unibas.ch/_tmp_files/sieverss/issue548-mas-issue548-base-issue548-v1-compare.html

If somebody wants to have a look at it, please go ahead:
https://bitbucket.org/SilvanS/fd-dev/pull-request/8/issue548/diff

Otherwise, I think that this task is accomplished.

We want to use the new task interface in merge-and-shrink.