Code is merged.

The code looks good to me.

Experiments look good to me. If the others are happy with the code, this is
ready to be merged.

Here is the new comparison report:
http://ai.cs.unibas.ch/_tmp_files/geissmann/issue524-v2-issue524-base-v2-issue524-v2-compare.html

The absolute report is updated.

Add a tag issue524-base-v2 to the default changeset which was last merged into
the issue branch.

The effect of this on performance are much larger than they should be. They
should be essentially imperceptible. I think the reason is that the branch
merged from default, but the experiment compares to an older base version from
default. Therefore, issue524-base vs. issue524-v2 includes several changes from
other issues in addition to the change from this issue.

People on the nosy list, what is our best practice for this? Update the
issue524-base tag, or introduce a new tag for the correct reference version?

Here are the results as a comparison report:
http://ai.cs.unibas.ch/_tmp_files/geissmann/issue524-v2-issue524-base-issue524-v2-compare.html

And the absolute report can be found here:
http://ai.cs.unibas.ch/_tmp_files/geissmann/issue524-v2.html

Yes, I would like to add lama-first so that we also exercise the code path that
uses this from the (new) synergy heuristics.

The code looks good to me. I think you could add lama-first to the existing 
experiment before running it.

Yes, please!

I merged the current default into this branch to be up-to-date. Maybe some can
have a look at the pull request.

Should I run some experiments for this issue?

Let me know if/when I should review the code. But if Jendrik has already
commented, it's probably a good idea to deal with his comments first. I hope
mine won't be contradictory. :-)

I left some comments on bitbucket.

Changed to shared_ptr for landmark graphs.

Here is the pull request:
https://bitbucket.org/cgeissmann/downward/pull-requests/5/issue524/diff