I suggest to close this one (and am doing it now, but feel free to change again
if you prefer) because the DTG produced by the preprocessor doesn't affect the
search any more.

Regarding changing the preprocessor output format, I have no strong opinion
whether to do it now or later. Whatever you prefer.

Since the DTGs from the prepropressor output are no longer used by the search
component, should we simply close this issue? Or should we first get rid of the
DTGs in the preprocessor output (issue26), which actually would be a five-minute
task (mainly updating the documentation)?

Neither issue434 nor this one has someone assigned to it. Anyone want to take
this on? If not, I can see if I can look into this.

Yes, I think we can close one of the two issues and add a pointer to the other one.

In the old issue, we mentioned the option to fix this at the same time as moving
the DTG generation into the search component, but in terms of running
experiments that we can easily analyze, it's perhaps better to fix it in the
preprocessor first (checking if that leads to a performance regression) and only
then migrate the code to the search component. That way, if performance is hurt,
we'll be able to assess whether it's because of the change to the DTGs or due to
the fact that they're computed in the search component.

(Adding Malte since I think that he should always be on the nosy list at least.)

No problem, so we can just close the old issue and refer to this new one.

Indeed this seems to be the exact same problem. Sorry, I was unaware this issue
existed already.

I think this issue is related (or even the same problem?): issue434

The description of the DTG in the preprocessor output is not always the same for
the same problem. Attached are two possible outputs for the problem
depot/pfile16. As far as I can tell the described DTG is the same but the
ordering of the transitions and transition conditions are different.
The problem seems to be very hard to reproduce (across 6 different repository
locations on 3 different machines i only had a different output in 1
repository), however I assume that the problem is the sorting of the vertices
vector in DomainTransitionGraph::finalize(). vertices is a
vector<vector<Transition> >, and the less-operator of Transition only compares
target, condition size and cost (if all 3 are equal, it is not defined which one
is smaller).