Merged and pushed.

Looks good to merge.

Good point, I removed the changes from sampling.cc (except for a wrong 
reference).

It looks like the changes to sampling.cc are unrelated to the title or
description of the issue. If we want them too, perhaps we should run experiments
on configurations that use this code. I'm not sure I understand the implications
of this change, e.g. if they could noticeably impact performance.

That makes sense. I changed the code accordingly. Can I merge this?

I think they should return strings. The considerations that apply here are no
different than for the proxies: if they return references, it is impossible for
these strings to be created on the fly. I think it makes no sense to have
AbstractTask::get_fact_name() return a const string & and FactProxy return a
string: this would contain the worst of both worlds. We would always copy
(because we access the information via FactProxy) but still preclude on-the-fly
generation (because FactProxy delegates to AbstractTask).

... and we could let them return by value.

I mean the methods 

AbstractTask::get_variable_name()
AbstractTask::get_fact_name()
AbstractTask::get_operator_name().

Currently, they all return references.

The patch looks fine to me. I'm not sure what you mean regarding AbstractTask;
which methods there do you suggest changing and how?

I created a pull request at

https://bitbucket.org/jendrikseipp/downward/pull-requests/57

The sie of the patch is very small so far, but the question is whether we also 
want to return by value from AbstractTask (instead of only from the proxy 
classes). I think that returning references there is fine.