Merged.

I agree that it would be good to have "general strategy" and I also think it could complicate the component interaction code since it is not clear how such an option object could be created and/or handled from different components. I won't be able to invest time in this in the near future, though.

Visually and stylistically, the diff looks good to me. If you have tested that it solves the problem, please feel free to merge.

Beyond this issue, I wonder if we should improve the current design where the defaults for these options interact in a somewhat complex way. It was flagged up as something that could potentially complicate the component interaction and Python library code because we will have multiple possible code paths related to default values. But we haven't looked at this deeply enough to say precisely how much of a complication this is or what would be helpful. Perhaps a "general strategy" (I think that's what we called this in the JAIR paper?) object would be the clean way to do this. Just bringing this up in case you are interested.

https://github.com/aibasel/downward/pull/276

Working on issue1191, I noticed that max_states, max_states_before_merge, and shrink_threshold_before_merge used by MergeScoringFunctionMIASM are not treated with handling default values as done in MergeAndShrinkAlgorithm. In the current version, leaving the values at -1, for example, results in assertion failures when attempting to compute the state limit before merging. 

When setting max_states (which is the usual/recommended way), this would not lead to errors which is why this was probably never a problem. It should still be fixed.