Merged and pushed.

Experiments look good to me, too.

I think this should be sufficient, yes. From my point of view, this change is 
purely a change of when things are computed. The amount of stuff that's computed 
should be unchanged.

I think the experiments cover all affected heuristics.

Are you sure the IPC14 benchmarks are sufficient? (If you are, it's fine with me.)

Did you test all affected heuristics?

I ran experiments on the IPC 2014 benchmarks (of course I forgot to exclude the ADL tasks in 
the optimal setting...):

http://ai.cs.unibas.ch/_tmp_files/seipp/issue591-v1-opt-issue591-base-issue591-v1-compare.html
http://ai.cs.unibas.ch/_tmp_files/seipp/issue591-v1-sat-issue591-base-issue591-v1-compare.html

While at it, I also revamped the common_setup.py interface a bit. As a result the interface is 
now closer to the base FastDownwardExperiment. This should make it easier to switch back and 
forth between the two for private and issue experiments.

Please run experiments first. I'll have a look at the code afterwards.

The first round of reviews is done, I guess. Malte, do you want to have a look at 
the code? Shall I run experiments?

https://bitbucket.org/jendrikseipp/downward/pull-requests/53

Please do the first round of reviews without me -- no time in the immediate future.

I had a look and made two comments.

I have made a pull request at 
https://bitbucket.org/jendrikseipp/downward/pull-requests/53. Can someone have a 
look, please?

Changed the title to reflect the primary task of this issue. We agreed to move 
the code from "initialize()" to the respective constructors. Once this is done, 
measuring the times of the different stages should be easier.

Currently there is still a use. We first have to change the current
"initialize()" structure. Can the two of you discuss the details between
yourselves? If you need further input from me, of course feel free to ask.

Remove output of actual search time, as there is no use for it anymore.