Merged.

not because of me

I fixed the comments in the pull request [1]. Do we need another round of
experiments for this?

[1]
https://bitbucket.org/flogo/downward-issues-issue486/commits/19802aff363e359245b9b1caed1aa5bf73a5ccac?at=default

I guess it's OK. The comment was mainly about how operator_costs is passed
through several methods that need to know the special meaning of an empty vector
without any documentation of these methods indicating this. So the knowledge
about the semantics of this operator cost vector is contained quite far from the
place where it is needed. I'm also not sure I like passing all operator costs to
build_abstract_operators if it only needs *one* operator cost -- I think it
would be cleaner to only pass in the one operator cost that the method is
interested in.

But I don't mind leaving this as is. Ideally we'll add a better, more general
mechanism for cost partitioning in the future based on the new task class.

Of course I can add a comment, but I don't understand what you mean by
"heuristic or PDB". The constructor already has a comment that explains the
parameter (line 97 of pdb_heuristic.h) but I guess you meant something else. The
comment for the constructor still uses the old parameter name which I just fixed
locally.

In all other respects very nice. :-)

It's a bit sad that the comments are gone; we used to have comments that explain
the purpose of the arguments (that it's there for cost partitioning) and explain
the default. Can we add something like this back? (One problem is that it's not
really clear which place is the best place to document it -- heuristic or PDB.
Maybe an explanation in the code that actually computes the cost and a reference
to that in the heuristic?) Feel free to merge, but I'd like it a bit better with
the comment back in. As a general rule, it'd be good if arguments to public
methods were either self-explanatory or explained.

Results [1] look good and Silvan already reviewed the code [2]. Can we merge this?

[1]
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue486-base-issue486-v1-compare.html
[2] https://bitbucket.org/flogo/downward-issues-issue486/branch/issue486

The PDB heuristic currently stores a copy of the cost function for cost
partitioned heuristics. This copy is only used during the heuristic construction
but persists for the whole life time of the heuristic object. We could use a
local variable in the constructor and pass it around in parameters instead of
storing it in a member variable.