If it doesn't fail in the issue211 branch, it's not relevant, but I seem to
remember it failed for me there recently (yesterday or two days ago).

Let me try again... looks fine now, wonderful! :-)

This happens in the default branch and not in the re-factored code of branch
issue211.

Is it still relevant?

Steps to reproduce:

$ make debug && ../translate/translate.py ../../benchmarks/mystery/prob07.pddl
&& ../preprocess/preprocess < output.sas && ./downward-1-debug < output --search
'astar(mas(count=1))'

=> gives:

[...]

Building atomic abstractions... unreachable: 1 nodes, irrelevant: 1 nodes
shrink by 1 nodes (from 2 to 1)
downward-1-debug: shrink_bisimulation.cc:144: void
ShrinkBisimulation::compute_abstraction(Abstraction&, int,
std::vector<__gnu_cxx::slist<int, std::allocator<int> >,
std::allocator<__gnu_cxx::slist<int, std::allocator<int> > > >&) const:
Assertion `h == abs.max_h + 1' failed.
Peak memory: 4384 KB
caught signal 6 -- exiting
Aborted


I should point out that this is an unsolvable task, so this is maybe a trivial
oversight.

Curiously, if I remove the count=1, I get a segfault instead of triggering the
assertion, even though count=1 should be the default. (?) It also looks like the
code doesn't complain when count is set to values less than 1, but it should.